NGDP Futures Targeting Is A Pretty Goofy Idea

NGDP Futures Targeting

A Pretty Goofy Idea

Oh, Hi Mark[ets]
Oh, Hi Markets

Expect The Expectations

First: I don’t intend to argue over the merits of targeting NGDP in general. For a background in that, you can start with Hall and Mankiw’s great paper from 1994, or check out Scott Sumner’s argument in the last issue of Foreign Affairs, or read some skepticism from Stephen Williamson and Noah Smith, or consider tangential affirmations of market-based solutions from Roger Farmer.[1] This post is only concerned with Sumner’s proposed “NGDP Futures Market” in which the Federal Reserve would list futures contracts linked to the Nominal Gross Domestic Product and set the monetary base as a result of the market’s activity.

At its core, NGDP futures targeting is about managing expectations. Rather than rely on forecast models, the Fed would set the monetary base (or the fed funds rate) according to a market’s expectations of the NGDP level at some point in the future. This “wisdom of the crowds” perspective comes from the remarkable accuracy that prediction markets have shown in forecasting future events. Initially this appears straightforward: the Fed lists a futures contract pegged at some level of NGDP growth, and if market participants, through buying or selling the contracts, express a view that the NGDP level will be higher or lower than the peg, the Fed will adjust its policy to bring those expectations in line with the target. However, when we look under the hood, I see a couple potential flaws that make me doubt how well this machine could run.

The Goofy Mechanism

In a typical zero-sum market, participants are compensated for taking a correct position with respect to the future value of an instrument or index. But in an NGDP futures market in which policy reacts to each buyer and seller, not only is accurate forecasting uncompensated, the policy mechanism is explicitly designed to punish expressed views. To see why, consider Sumner’s automated policy suggestion:

Eventually, the buying and selling of government bonds could even be automated, with every “long” or “short” purchase on the futures market triggering a corresponding open-market operation. For instance, each $1 purchase of a long position in a nominal GDP futures contract might trigger a $1,000 open-market sale by the Fed, while a $1 purchase of a short position would trigger a $1,000 open-market purchase by the Fed. Investors would effectively be determining the size of the monetary base.

Breaking down the mechanics you get:

  1. You believe NGDP will be lower than the Fed’s posted market of x% at expiration
  2. So you sell a futures contract to express this view
  3. The Fed responds by increasing the monetary base
  4. Thus raising expected NGDP against your view

So… by selling that futures contract you have just raised the expected settlement value against yourself. It’s like if selling a futures contract in copper triggered a mechanism which automatically lowered the existing quantity of copper. Expressing your view immediately causes your view to be less right. I’m not suggesting that people don’t respond to prices all the time — e.g. a farmer seeing a high price for future grains and subsequently planting more of them. But in general, a well-functioning market shouldn’t have an infinitely liquid counter-party triggering automatic causal effects against every position.[2]

I think such a mechanism not only defeats the purpose of price discovery, but in Sumner’s market it might lead to other hilariously manipulative scenarios. Consider: because the Fed is pegging the price of the futures contract and buying/selling infinite quantities, a smart player could load up on assets whose prices would be affected by a large expansion of the monetary base, then sell a tremendous amount of NGDP futures to trigger open market purchases, then sell the aforementioned assets and subsequently repurchase the NGDP futures at the same price for a tidy profit.

Revisions are Non-Trivial

As a future is a derivative of the price of some underlying commodity or value of an index, a well-designed contract is linked to a contemporaneously measurable value that does not suffer from ex post revisions. For the popular E-mini S&P 500 Futures traded on the CME, the value at which the contracts settles at expiration is based on the opening prices of the components in the underlying index. These prices are real values that cannot be modified by someone a month later after “more complete data” comes in. Even now as it looks like the CME intends to eventually begin trading in Bitcoin futures, the first step was to construct a “Bitcoin Reference Rate (BRR), which will provide a final settlement price in U.S. dollars at 4 pm London time on each trading day.” Clearly, the integrity of a properly designed futures contract is dependent upon reliable and accurate values of the underlying product or index.

In contrast, measurement error overwhelms the reliability of GDP releases, and the numbers posted by the Bureau of Economic Analysis are heavily revised for a long period of time afterward. The BEA itself states that its estimations are good for +/- 1%. Later this month the BEA will release its “annual revision of the national income and product accounts, covering the first quarter of 2013 through the first quarter of 2016.” In the most recent release, the fourth quarter 2015 GDI number was revised up from 0.9% to 1.9%. This implies the measurement errors are on the same order of magnitude as the estimates! I argue that this is not trivial. But Sumner thinks we can ignore revisions:

The payoffs can be based on the first announcement, as long as the initial estimate is unbiased. Any longer-term revisions in the definition of GDP, such as adding the underground economy, should be addressed by adjusting the policy goal.

This isn’t very satisfying. Of course if initial estimates are unbiased in the long run then in theory all the errors should eventually come out in the wash. But this conveniently ignores a reality where people have real money on the line. In Sumner’s proposal, an interested firm or trader who spends resources to predict the correct value of the GDP number can suffer an endemic loss due to measurement error. There are a few possible consequences of this: firms simply won’t want to trade such a contract; considerable resources could be spent on trying to predict the initial volatile estimate (rather than the true value); or firms could account for the uncertainty by trading less of the contract or hedging against something else. Each of these scenarios implies that prices will not reflect actual expectations, and the latter two are biased toward the noisy process.

As a result, I predict there would be little to no volume in any NGDP futures contract.[3]

To test this claim we can look at the history of an instrument with similar revision problems. Over 10 years ago, the CME had been launching futures contracts linked to the monthly U.S. non-farm payroll number (aka the BLS employment situation). It’s the biggest economic number in the world, even moving European markets more than any releases specific to Europe. Surely people would want in on this action? Well no. The revisions to that number are also on the same order of magnitude as the releases. At the end of 2011, the CME delisted all of them with this release note:

Effective Monday, December 5, 2011, CME is delisting Nonfarm Payroll futures and options. This notice informs contract users of this delisting. Nonfarm Payroll futures and options have not traded since inception and there is no open interest.

Emphasis mine.

Let It Be

There is some decent market rationale for why things are the way they are:

If the Fed wanted to manage expectations by being active in futures markets, it could have been doing so for a long time already. There exists a fairly liquid and actively traded futures product linked to the overnight rate called 30 Day Fed Funds Futures. The problem here is that if it were known that the Fed were directly trading in these products, everyone else would leave. No one wants to place futures bets against the monetary authority in an instrument that it can directly influence on a whim. But nothing is stopping the Fed from using that implied forward yield curve as guidance for what the market is thinking. They probably already do.

Finally, creating a listed futures contract nowadays is a trivial procedure.[4] If the listing of NGDP futures in general were really such a super cool awesome idea, we’d probably have seen it already. There are futures contracts for regional home price indices. There are futures contracts for average monthly temperature changes. There once were futures contracts for the number of hurricanes that make landfall in the US in a given season. So why not GDP futures? The competition between exchanges and the incentive to be the first to market is just too great for this to have been overlooked for so long. Have you been paying attention?


With big thank yous to professor Mark Witte for helping me develop this idea and to Noah Smith for proofing and critiquing.


Update 7/10/2016: Read Mike Sankowski

Mike Sankowski is a smart dude writing over at Monetary Realism whose professional work actually involved designing futures contracts. He wrote some similar thoughts from a slightly different perspective on the NGDP futures market four years ago to this day. And then some more. So if you don’t find my arguments convincing, or if you need a stronger background in what’s going on, read Sankowski. He’s good.

Update 7/19/2016: Sumner’s Response Falls Flat

Sumner issues this response.

I can appreciate how much of a long uphill battle Sumner has fought for NGDP targeting and the NGDP futures idea. He deserves a lot of credit for championing the idea of NGDP targeting to where it is now in the main stream collective consciousness of academics as well as a broader econ-savvy population. Without his considerable efforts, I probably wouldn’t have been following this subject over the past six years or so. That said: his incoherent response shows a failure to understand my criticism and possibly a failure to understand his own market design.

In true Sumnerian fashion, he begins with an off-hand remark about how I ignored/didn’t read his proposal. Any long time follower of his, like me, knows that this is Sumner’s standard opening move for responding to all criticisms of NGDP targeting. I’ve read it all; it’s still goofy. (though not as goofy as the time he called Arctic Monkeys a one-hit wonder)

Let’s start here:

He wonders why NGDP futures would be such a good idea, given that the private sector hasn’t already created such a market. Perhaps that’s because the private sector is not legally allowed to do monetary policy.

Oof. This is embarrassing. Sumner attempts to imply that we haven’t seen a private sector futures contract linked to NGDP because it would necessarily “do monetary policy” which the private sector cannot. It’s a gross non sequitur and completely ignores my point. In the main piece, you’ll see that there are no fundamental or market structure issues preventing the creation of an NGDP futures contract. I use the unpopularity of the former unemployment-linked contracts as an analogous example of why his market might have problems gaining traction. Dressing up a futures contract as “monetary policy” does not make it any less of a futures contract.

After not realizing he was making the case for me for why pegs are a bad idea (did an economist really just extol the virtues of gold standard pegging?), Sumner offers this mess:

Under NGDP futures targeting, manipulation would be even more difficult. If you sold the asset back immediately, it would not significantly impact monetary policy. Now consider what would happen if you held the NGDP futures until maturity. If you bought NGDP futures to trigger a contractionary monetary policy, and it succeeded, you would lose money on the NGDP futures (unlike the previous gold example where you break even on gold.)

To see why this is wrong, it’s helpful to abstract a bit away: if the price of Instrument A is pegged, and buying/selling Instrument A automatically affects the price of Instrument B, then you can manipulate the price of Instrument B without losing money on Instrument A. If trading NGDP futures contracts (Instrument A) automatically triggers open market operations, then the price of things analogous to Instrument B will be affected in real time. In practice, this can be done very quickly such that profits can be made and the unlimited supply of NGDP contracts can be re-sold/bought at the original price. Perhaps Sumner doesn’t understand the ramifications of his own automated monetary policy proposal.

He offers little consolation:

In that case, you need even bigger profits on your side bets. The problem is obvious—other “manipulators” will be taking the opposite strategy. If they respond to you by selling NGDP futures, moving monetary policy closer to the correct level, they will breakeven on the NGDP futures and profit (at your expense) on their side bets in other markets.

Facts not in evidence. The existence of many players does not preclude manipulation. Further, in Sumner’s proposal, the Fed is the only market maker. All players must take the posted liquidity from the Fed. Therefore, it is not necessary for other people to be taking the other side of the bet at any time. Again, this shows a very poor understanding of the market structure.

Lastly, I’m glad Sumner was able to generate some interest by subsidizing his fanboy NGDP futures project that didn’t have a price peg. But this post clearly deals with the pegged price proposal found in this paper, and the automated monetary policy rule he suggested.

With love, “Zacharay”

I’m at the Montreal “Just For Laughs” festival this week but wanted to acknowledge ongoing contributions:

  • Sumner objected to my update
  • Noah Smith made a great post breaking down my argument into more palpable chunks
  • Sumner hits both of us back on Econlog
  • Noah updated his original post with a response to that
  • Brian Romanchuck dissects the futures convertability proposal (spoiler: it too is silly)
  • Cullen Roche discusses the challenges between theory and real market behavior

The gist is that Sumner waves the Efficient Markets Hypothesis flag over his convoluted mechanism to claim that informed traders will trade even when it necessarily moves the expected settlement against their information while also disregarding the problems of noise and revisions. He also uses EMH to wave away my example of how manipulation can occur in such a rule-based regime. Noah correctly points out some of the flaws with that magical thinking and gives a clear example using price pegs.

I don’t have anything to add right now because “EMH, bro” isn’t a very interesting or convincing position to me.

At this point, if Sumner is so certain that none of us “get it” then he should probably acknowledge that he just isn’t very good at using words. Perhaps he should spend some time formalizing his proposal so that his garrulous treatments aren’t such easy targets. Adding kludge after kludge doesn’t seem to be working out too well for him.

P.S. I appreciate the four or five people who got the reference in the picture at the top of this post. I’ve heard that there is a video somewhere on the Youtubes with me and a few former coworkers all dressed in tuxedos with Mr. Wiseau himself.

Footnotes    (↵ returns to text)

  1. Professor Farmer has a new book coming out this fall called “Prosperity for All: How to Prevent Financial Crises.” (link) It looks like it’s going to be very good.
  2. To be fair, Sumner isn’t the only one suggesting paradoxical market mechanisms these days. Robert Shiller and Mark Kamstra have proposed that government’s finance spending by selling financial products which pay dividends based on GDP, called “Trills.” (link) But this also has a potentially weird design. Consider the scenario: a person is willing to buy a Trill at some price which factors in expected GDP values, then the government sells that person a Trill. In a normal market, selling activity typically forces the price downward. But if the government uses the money from Trills to finance spending, that spending becomes a part of the GDP number, which would increase the dividend, then people would upwardly revise their expected dividend and then raise the price they would pay for Trills. The obvious counter-argument is that the government spending generated through the revenue of selling a Trill will only effect one year’s GDP number. Still, the implication is interesting to consider.
  3. Obviously with the exception of the possible manipulation scenarios.
  4. They’re so easy to create. But it’s very difficult to get people to actively trade them. More than 80% of listed futures contracts will have zero volume on any given day.